register IPython's eventloop integration in plt.install_repl_displayhook

[minrk]

Registers eventloop integration for IPython on setup, avoiding hangs when IPython hasn't been told about matplotlib prior to plotting.

The IPython kernel (notebook, qtcosole) has always needed this to avoid hangs if plotting happens prior to %matplotlib magic, but it was less important for terminal IPython < 5.0. Terminal IPython 5.0's adoption of prompt_toolkit means that it has the same eventloop requirements of the kernel, so it's now important in both contexts to avoid hangs.

cc @tacaswell

[tacaswell]

What does this do with backends that ipython can not map to a name?

[minrk]

Nothing. Since it's a get from a dict, gui_name will be None and it will take no further action, which can be interpreted as either:

1. IPython doesn't know how to integrate with the eventloop for the given backend, or
2. it doesn't need to (e.g. Agg)

[tacaswell]

🐑 right. Thought that backend2gui.get was something much more complicated.

[janschulz]

Closing and reopening to run with miktex changes on appveyor to see if that lets the appveyor tests pass.

[afvincent]

For information, for the moment mpl is quite unusable with IPython 5 when using an interactive session.
For example, on my computer (Linux OS), both with mpl 1.5.1 and 2.0.0b2.post1762+gcf24e05, when doing

import matplotlib.pyplot as plt
plt.ion()
fig = plt.figure()

I have to use Ctrl + D to make the new figure to appear… Morevoer, if I answer n to the prompt, and then want to close the figure window, I have to force its halt (through my DE, Xfce) and IPython is also brutally halted…

The issue doesn't not occur when I don't use an interactive session, and call plt.show() instead.

[tacaswell]

@afvincent just run %matplotlib, this has always been the recommendation of both mpl and IPython. That it was ever working with out the magic was an implementation detail of the GUI toolkit integrations.

[tacaswell]

The windows failure looks tex related.

[afvincent]

@tacaswell : thank you, with the magic command, everything is fine. It seems that I never looked carefully enough to the docs to notice the lines about using %matplotib…

[janschulz]

The windows failure looks tex related.

Seems that the fix in master isn't picked up here :-/

[tacaswell]

pep8 that is definitly not @minrk 's fault, merging as-is.

[tacaswell]

backported to v2.x as a61f763

[tacaswell]

cherry-picked back to 1.5.x as 44c206b